By LARRY SWEDROE
Academic research has found that valuation metrics (for example, the earnings yield (E/P) or the CAPE 10 earnings yield) and valuation spreads have information in terms of future returns. The higher the earnings yield, the higher the expected return; and the larger the spread in valuations between growth and value stocks, the larger the future value premium is likely to be. What’s more, this relationship holds across asset classes, not just stocks.
This relationship should be expected if the cross-sectional book-to-market spread is a proxy for the price of risk. When the price of risk increases, the cross-sectional spread in risk premiums (and consequently, book-to-market) widens, and the expected premiums to factors such as market beta, size and value also increase.
For example, the November 2019 study , by Fahiz Baba Yara, Martijn Boons and Andrea Tamoni, found that valuation spreads provide information about future expected returns. The authors found that “returns to value strategies in individual equities, commodities, currencies, global government bonds and stock indexes are predictable by the value spread… In all these asset classes, a standard deviation increase in the value spread predicts an increase in the expected return to the value factor that is the same order of magnitude (or more) as the unconditional value premium itself.”
Jim Davis’ 2007 study, , also found that book-to-market ratio spreads contain information regarding future returns. However, he also found that style-timing rules do not generate high average returns because the signals are “too noisy” — they don’t provide enough information to offer a profitable timing signal.
Thiago de Oliveira Souza provides further evidence on time-varying factor premiums with his May 2020 study, in which he investigated whether the variation in the overall market price of risk induces similar variation in factor risk premiums such as those of Fama-French. His data series on the factors of size, value, profitability and investment covered the period June 1963-August 2018, and for market beta the period 1926-August 2018. The following is a summary of his findings:
There are common variables that forecast all factor premiums (except the profitability premium) and provide evidence that individual stock premiums also vary in coordination with the market price of risk.
Increases in the cross-sectional book-to-market spreads significantly forecast increases in one-month-ahead premiums for all except the profitability factor.
Increases in the investment (profitability) spreads marginally forecast increases in the investment (profitability) premiums.
De Oliveira Souza concluded that his finding supports the risk-based explanation of all but the profitability premium, but cannot be taken as strong evidence against the risk-based explanation of the profitability factor.
Antti Ilmanen, Ronen Israel, Rachel Lee, Tobias Moskowitz and Ashwin Thapar contribute to the asset pricing literature with their study , published in the Fourth Quarter 2021 issue of the Journal of Investment Management, in which they examined four prominent factors (value, momentum, carry and defensive) across six asset classes over a century. Their data sample began in 1926 and covered individual equities, equity indices (U.S. and international), government bonds, currencies and commodities. Individual equity data from 21 international markets began in 1984. Individual stock universes comprised 90 percent of the total market capitalisation of each market (they excluded the smallest stocks). Following is a summary of their findings:
The premia were all positive for each factor in each asset class and the majority were statistically significant. The premia for each factor were also positive in the majority of decades, with only a handful of instances of decade-long underperformance for a factor.
When applied across all asset classes, the Sharpe ratios for value, momentum, carry and defensive were 0.53, 0.64, 0.57 and 0.68, respectively.
Sharpe ratios were generally of greater magnitude in stock selection than in other asset classes.
When pooling across asset classes, the t-stats of the mean returns ranged from 5.1 for value to more than 6.6 for defensive—rejecting a null hypothesis of the factors being uncompensated. The multifactor portfolio that combined all four factors across asset classes had a Sharpe ratio of 1.46 with a t-stat of 14.2, indicating large diversification benefits from combining different factors and different asset classes (a t-stat of 14.2 would require more than one trillion random searches to be generated purely by chance).
There is meaningful time variation in factor risk-adjusted returns that appears unrelated to macroeconomic risks—there were no significant exposures of the factor returns to economic activity or news, although they did find some variation in the risk and correlation structure of the factors to the economic environment.
There was little evidence that factors varied with interest rate environments in a manner purported by duration models.
The return per unit of risk varied significantly over time even for a portfolio that was diversified across factors—the diversification across factors does not completely ameliorate the changing risk of each factor over time.
Correlations between the factors varied over time, contributing to the variation in risk of the multifactor portfolio. However, there were no periods when correlations all became large enough to severely reduce diversification benefits, and there was no apparent upward trend in correlations.
There was relatively modest predictability that likely fails to overcome implementation frictions—there was weak and inconsistent evidence for factor timing. The most consistent results came from using valuation spreads and inverse volatility to time factors.
There was little evidence of arbitrage activity influencing returns.
Their findings led the authors to conclude: “We find robust out-of-sample evidence of factor premia, rejecting that they are the result of spurious data mining, but find that overfitting biases may contribute to a significant decline in the out-of-sample efficacy of these factors.”
They added: “Our findings at least soften the conclusion that arbitrage activity has led to a reduction in factor return premia given that we find no evidence for it across a variety of factors, asset classes, and time periods.” And finally: “The case for adding factor timing to an already diversified multifactor portfolio is tenuous in practice. Despite looking at a plethora of timing strategies, methodologies, and signals, we find modest evidence of out-of-sample factor timing. Accounting for increased turnover and trading costs associated with factor timing, the net of cost returns to timing are likely de minimis. On a more positive note, despite limited ability to profit from factor timing in real time, we find significant conditional return premia associated with common factors across diverse asset classes.”
The body of evidence provides strong support for the persistence and pervasiveness of the value, momentum, carry and defensive factors. It also demonstrates that the valuation spreads do contain valuable information as to future premia returns. With that said, the study by Ilmanen, Israel, Lee, Moskowitz and Thapar also provides evidence that suggests investors are likely best served by avoiding deviating from a long-term static allocation to these factors through tactical factor timing.
© The Evidence-Based Investor MMXXIV. All rights reserved. Unauthorised use and/ or duplication of this material without express and written permission is strictly prohibited.